fayerman-source stops Mike from quietly writing privileged material to disk

In the original code, every interaction with the AI engine - the prompts going in, the document context attached to them, and the model's answers coming back - was written to a plaintext file on the server. It happened on every call, with no off switch, and the file simply kept growing for as long as the deployment ran.

fayerman-source removed it. The disk-writing is gone entirely, and the only logging that remains is an explicit debug mode that stays off unless someone deliberately turns it on. The change also flags that other forks which extended the AI layer likely carry the same hidden logging - worth a look if you're auditing one.